fuxicodex/Fuxi
FuXi is a terminal-based AI agent that edits code and runs commands on its own
FuXi is an AI coding agent that runs inside your terminal, reading code, editing files, running shell commands, and driving tools — all packed into a single self-contained binary. It routes each request to a cheap or powerful model depending on task complexity, with automatic failover across providers. No separate runtime installation is required.
What it does
- Instead of just answering questions, it runs a think-act-verify loop: reasoning about a task, acting with tools, checking results, and repeating until something like a failing test is actually fixed and verified.
- Every request is scored for complexity and routed to the appropriate model tier — cheap models for simple tasks, powerful ones for hard tasks — with automatic failover if a provider fails.
- It ships with 50+ built-in tools (file editing, shell commands, code search, web fetch, diagnostics, Jupyter) plus parallel sub-agents, all in one binary.
- Shell commands pass through an AST-based safety classifier and rule checks before running, backed by fine-grained permissions and audit logs to keep autonomous actions under control.
- Users can bring their own API key for providers like OpenAI-compatible endpoints, Gemini, or Bedrock/Vertex, or simply log into a FuXi account to use managed models.
Why it matters
For developers juggling multiple LLM providers and trying to balance cost against capability, a tool that automatically routes tasks by difficulty offers a practical way to save money without losing quality. That said, the repository does not publish scores on benchmarks like SWE-bench, and the core source code is proprietary, so performance claims need to be verified independently.
Terms in this repo
- TUI · a text-based interface that lets you navigate menus and screens inside a terminal
- MCP · a standardized protocol that lets AI models connect to external tools and data sources
- failover · automatically switching to a backup service when the primary one stops responding
- AST safety classifier · a checker that analyzes a command's syntax structure to judge whether it's risky before running it
- static binary · a single executable file that runs without needing separate runtime installations or libraries
Repository description (English)
edit code, run commands, and drive tools, with cost-aware routing across LLM providers.
Open on GitHubTrending repos
- vorssaint/vorssaint-utilsOne free menu bar app replaces a dozen paid Mac utilities
- Alishahryar1/free-claude-codeA local proxy that lets coding AI agents run on 49 free or cheap model providers instead of one paid service
- freestylefly/awesome-gpt-image-2A library of 532 reverse-engineered prompts that turn GPT-Image2 into a predictable image-making tool
- block/buzzAn open-source workspace where humans and AI agents chat, code, and review in the same rooms
- NousResearch/hermes-agentNous Research's Hermes is an AI agent that gets smarter the more you use it
- virgiliojr94/book-to-skillA tool that turns technical book PDFs into on-demand reference skills for AI coding agents
- VoltAgent/awesome-agent-skillsA single hub collecting over 1000 'how-to' manuals that make AI coding assistants act like experts
- anthropics/claude-plugins-communityA shared shelf where anyone's Claude add-ons get listed for install